The Blowback Inherent to Network Analysis Kill Lists

As I noted Gregory McNeal has a fascinating series of posts at Lawfare on how the government develops kill lists (this post even has a mock-up of a Kill List baseball card). But I find this post, which describes how the Kill List makers use network analysis to pick and choose whom to kill, the most interesting. It implicitly reveals one of the most fundamental problems with the way we’re doing drone targeting.

McNeal describes how the government uses network analysis to find the most crucial people in the functioning of a terrorist network. Those people, he explains, may not be the most senior or public members of the group.

Networked based analysis looks at terrorist groups as nodes connected by links, and assesses how components of that terrorist network operate together and independently of one another.  Those nodes and links, once identified will be targeted with the goal of disrupting and degrading their functionality.  To effectively pursue a network based approach, bureaucrats rely in part on what is known as “pattern of life analysis” which involves connecting the relationships between places and people by tracking their patterns of life. This analysis draws on the interrelationships among groups “to determine the degree and points of their interdependence.” It assesses how activities are linked and looks to “determine the most effective way to influence or affect the enemy system.”

[snip]

Viewing targeting in this way demonstrates how seemingly low level individuals such as couriers and other “middle-men” in decentralized networks such as al Qaeda are oftentimes critical to the successful functioning of the enemy organization. Targeting these individuals can “destabilize clandestine networks by compromising large sections of the organization, distancing operatives from direct guidance, and impeding organizational communication and function.” Moreover, because clandestine networks rely on social relationships to manage the trade-off between maintaining secrecy and security, attacking key nodes can have a detrimental impact on the enemy’s ability to conduct their operations.

McNeal’s description of the role of network analysis is not entirely new. We’ve seen hints of it in the drone speeches made by various officials. But the description greatly fleshes out what the government thinks it is doing when it engages in pattern of life analysis.

From there, McNeal explains that a range of outsiders — NGOs, journalists, even family members — may not be able to see what the network analysts privy to this magic information can see, the crucial role someone has in a terrorist network.

Thus, while some individuals may seem insignificant to the outside observer, when considered by an analyst relying on network based analytical techniques, the elimination of a seemingly low level individual might have an important impact on an enemy organization. Moreover, because terrorist networks rely on secrecy in communication, individuals within those networks may forge strong ties that remain dormant for the purposes of operational security. This means that social ties that appear inactive or weak to a casual observer such as an NGO, human rights worker, journalist, or even a target’s family members may in fact be strong ties within the network. Furthermore, because terrorist networks oftentimes rely on social connections between charismatic leaders to function, disrupting those lines of communication can significantly impact those networks. [my emphasis]

Even assuming the software that lays at the core of network analysis provides better knowledge than the deeply embedded understanding of those more familiar with the culture in question (and for a robust view on that, see Haley Barbour on steak power lunches), there is a serious problem with this result.

The most complete description of the network analysis that lies behind our drone killing makes it clear that members of a target’s own community may not understand why he was targeted.

Adam Baron, describing the aftermath of the drone killing of the Yemeni Adnan al Qadhi in Beit al Ahmar last year, shows what happens when members of a target’s community don’t understand why he was killed.

Few here dispute Qadhi’s open sympathy toward AQAP. After all, the target’s house, modest compared to nearby fortress-like compounds, sticks out because of a mural on one side that shows al Qaida’s signature black flag.

But his relatives and associates say there’s more nuance to Qadhi’s story. While he was labeled as a local leader of AQAP after his death, as recently as last winter he’d participated on a team that mediated between the government and AQAP-linked militants who’d seized control of the central town of Rada. The scion of a prominent local family who still held a position as an officer in the Yemeni military, Qadhi had refused to take part in the fighting, relatives said. They said he stayed home even as other AQAP militants carved out a base in the southern province of Abyan.

“He may have supported al Qaida, but he wasn’t taking part in activities,” said Abdulrazzaq Jamal, a Yemeni journalist and analyst who met with Qadhi shortly before his death. “There were connections, but there wasn’t perceptible tangible support.”

While Qadhi appeared to make little secret of his extremist ideology, his relatives said the strike against him came as a total shock. There had been no indication that he was a potential drone target, they said. Had they known he was considered such a high-value target, they claimed, they would’ve assured his cooperation with the authorities.

[snip]

His neighbor, Mohamed Abdulwali, took a break from repairing a water canister to chime in: “Any action has a reaction. Any violence will breed violence.” [my emphasis]

It’s not that Qadhi’s neighbors didn’t know about his support for AQAP. But they had a very different understanding of what kind of threat he posed — particularly given his role in mediating between locals and AQAP and his decision not to engage in hostilities — then the network analysts in the US who ordered up his death.

And that different understanding made the US strike illegitimate in their view, a perceived violation of rule of law. It led to open calls for a violent response.

In short, it converted an otherwise neutral community into one opposed to the United States.

If network analysis results in killings that local communities do not understand and therefore consider illegitimate, it will lead to us losing the political battle for hearts and minds.

There are more potential problems that come from network analysis killing. For example, unless the analysts are also doing network analysis of the surrounding community, they may miss the role a person — and Qadhi is a perfect example — might play in persuading locals to turn against al Qaeda. That is, killing someone like Qadhi may rule out what we did with the Sons of Iraq, effectively undercutting a really violent insurgency by buying the loyalty (or perhaps renting, as this violence is returning now) of key leaders within the insurgency. Aiming to kill the key figures in the network may not be the most efficient way of achieving peace and stability, even if it allows Administration figures to boast about stomping out the enemy.

But at its core, it’s the (claimed) asymmetric understanding of this network that makes this kind of killing so stupid. Drone killing that presumes a special knowledge about individuals’ roles in a terrorist network — but doesn’t share it with the people whose sympathy we must have to win this fight — is bound to backfire.

Update: McNeal reminds me that network analysis involves human analysts assisted by software, not just software. It’s a fair point. To be clear, though, I’m not dismissing the value of network analysis (though I question how good our HUMINT going into it is). I’m suggesting that information asymmetry makes it really dangerous to use.

Marcy has been blogging full time since 2007. She’s known for her live-blogging of the Scooter Libby trial, her discovery of the number of times Khalid Sheikh Mohammed was waterboarded, and generally for her weedy analysis of document dumps.

Marcy Wheeler is an independent journalist writing about national security and civil liberties. She writes as emptywheel at her eponymous blog, publishes at outlets including the Guardian, Salon, and the Progressive, and appears frequently on television and radio. She is the author of Anatomy of Deceit, a primer on the CIA leak investigation, and liveblogged the Scooter Libby trial.

Marcy has a PhD from the University of Michigan, where she researched the “feuilleton,” a short conversational newspaper form that has proven important in times of heightened censorship. Before and after her time in academics, Marcy provided documentation consulting for corporations in the auto, tech, and energy industries. She lives with her spouse and dog in Grand Rapids, MI.

20 replies
  1. P J Evans says:

    ‘Network analysis killing’ reminds me of M A Foster’s ‘Morphodite’: someone who can look at the world and figure out who needs to die in order for necessary change to happen – and it’s usually an apparent nobody. (In the end, the change requires that the Morphodite itself die.)

  2. emptywheel says:

    @What Constitution?: I’m not dismissing the value of network analysis entirely. But I was serious when pointing to Haley Barbour’s post, which basically says HUMINT isn’t as good as SIGINT for mapping out power within a community, and we’re not using enough HUMINT w/drone strikes.

    I’m saying that used as it is, it will backfire. Because drone strikes that seem inexplicable to locals by definition will not help us win legitimacy.

  3. scribe says:

    You note: “In short, it converted an otherwise neutral community into one opposed to the United States.”

    From the perspective of the analysts and their supervisors, all the way up the chain, this is not necessarily a bad thing. After all, continued conflict means continued employment and career advancement (e.g., bigger budgets for bigger and better toys of all sorts).

    It’s the counter-terra analog to a problematic question I pose to anyone demanding an immediate legalization of presently-illegal drugs: whaddya going to do with all the now-suddenly-unemployed cops (who’ve devoted their entire professional lives and, thereby, all their education and experience to chasing and arresting dope-oriented malefactors? They’re going to want to stay cops, not the least because they are best at being cops. And that means they’ll need something new to arrest people for.

    Refining targeting for targeted killings (which is a constant process) just means these analysts (and software developers) get better at deconstructing societies violently. And they’ll do it in a way that guarantees their continued employment. They got kids who need braces and college funds filled.

  4. What Constitution? says:

    @emptywheel: Understood. I completely agree that if the “public face” on these targetings is “trust us”, there’s no reason to expect anyone who lives under these drones or reports on the drone strikes to presume legitimacy to the targeting decisions. Maybe “we know it when we see it” isn’t as apt a description of the operating assumptions regarding the importance of being understood; maybe the better one comes from G. Gordon Liddy’s wall plaque: “When you’ve got them by the balls, their hearts and minds will follow.” Or was that some other Nixon stooge?

  5. lefty665 says:

    @emptywheel: Might be useful to refine “sigint”. Specifically, the quotes seem to indicate they are mostly working with “traffic analysis”, who is talking to whom, a subset of sigint, rather than the content of the comm.

    The Barbour link illustrates potential problems with sigint content. The data may be relevant, but not have much to with what actually happens. The right traffic may not have been intercepted (wrong restaurant/wrong meal), or that conclusions about what comm was dispositive are wrong.

    Humint is not necessarily much of a solution. It can, and does, suffer from all the defects of sigint plus some. With intercepted signals, where the targets do not know they are being copied, you’re not likely to have a person blow smoke at you. It’s the essence of a long term argument between CIA/NSA.

  6. Tom Allen says:

    I wonder to what extent they consider that elimination of one “node”, however critical, can be compensated for by the addition of multiple new nodes. (“For every terrorist you kill you create a dozen more”, e.g.) Presumably they’re treating these groups as scale-free networks, and think that by eliminating major hubs (critical nodes) they can split one large network into several isolated ones. But that would have to be done fairly quickly, or the links can be reforged. (Example: how many #3 Al Qaeda leaders have been killed so far?)

    And this all assumes, even given perfect information, that their models are relatively accurate — which, considering how new network theory is, is a dangerous assumption to make.

  7. Saul Tannenbaum says:

    Network analysis will give you a map of relationships and, as cartographers will tell you again and again, a map is not the territory.

  8. SpanishInquisition says:

    “In short, it converted an otherwise neutral community into one opposed to the United States.”

    But wouldn’t network analysis itself target neutral communities – like family members, schools, etc that explicitly aren’t involved in terrorism and might not even know what’s going on? I would expect many things could converge at schools, marketplaces, family homes, etc, but the majority of people at those places would be entirely ignorant of what was going on. I’m particularly wary of this aspect given the Obama administration’s stance on targeting things like first responders and funerals and basically saying that these deaths are proof itself of the targets’ guilt…I believe those strikes are based on network analysis.

  9. Peterr says:

    Network analysis may be good at identifying otherwise hidden relationships, but the jump to the use of drones and a kill list is problematic on many many levels, from the moral to the practical.

    Focusing solely on the practical . . .

    People function within various social networks, which may or may not overlap. Killing someone removes them from ALL these networks, not only the AQ network. This person’s death may leave the AQ network confused and broken in terms of their functionality, but as Marcy notes, it potentially leaves other networks angry where they were neutral in the past. Thus, members of these second and third networks now may be inclined to become active members of the first, including becoming suicide bombers and martyrs for The Cause.

    Killing someone like this disrupts a network; it does not dismantle it or render it inoperative permanently. The network may find someone new to fill that position, or may adjust itself around the broken node. At best, taking out one person is a tactic, not a strategy.

    Consider, for example, another tactic: the Shakedown.

    Hello, Mr. A. Nice home you’ve got here. Nice job you’ve got there. Nice family. It would be a shame if anything happened to them. You know, like if you were to meet with a sudden accident. That would be a real shame. . . .

    You know, we’ve learned that you have some not-so-nice friends. We don’t want to burden you with how we know, but we know. We know about Mr. X, Mr. Y, Mr. Z, and the rest. Hanging out with them could be dangerous to your health. . . . . Have a nice day, and make sure to kiss your wife and kids when you get home. If you get home.

    Immediate Result: Mr. A now knows he’s been identified. Does he continue as before? Does he run and hide? Does he tell his friends X, Y, and Z? Do they run and hide? Does he loudly and publicly proclaim the threat he’s received, and likewise his innocence?

    (Note: killing Mr. A with a drone also lets X, Y, and Z know that A was identified, so don’t complain that The Shakedown reveals something to the network.)

    Different Mr. A’s will react differently, but if disruption is your goal, you’ve accomplished it. And chances are, you’ve accomplished it with far less collateral damage in terms of inflaming other non-AQ networks of relationships.

  10. P J Evans says:

    So that’s why NYPD was surveilling all those food places. And why the national agencies watch falafel joints.

  11. orionATL says:

    i am deeply skeptical when a law, police, or military bureaucracies turn to using theory and techniques from the social sciences. no doubt network analysis of the sociological sort has improved since the mid-80’s when i worked with it, but putting it to use to determine importance in an organization and to kill people strikes me as likely placing a greater burden on the theory and the analytical techniques than they can carry.

    though it’s not the same, we saw the folly of using social psychology and learned helplessness to put an asceptic, scientific face on torture.

    results from “public opinion” surveys are often used for decision making despite the well-know flaws built into the dsta collection.

    i believe the u.s. dep’t of justice has introduced some sort of social science measure into material support for terrorism trials purporting to measure the tendency of a jihadist supporter to influence others to take up that path.

    the calculated social isolation techniques used by the military on jose padilla and bradley manning may have been effective uses of the science, but they were a per ersion of it.

    and of course we have the fbi’s “personality/psychology” lab at quantico that would likely have been involved in the bruce ivans harrassment and suicide, and posdibly in the aaron swartz affair.

    i would ecpect that this use of network analysis is tremendously useful for the eliminate-the-terrorists bureaucracies because it gives easily understood actionable results – get shorty! he’s the guy.

    i would think some readings of papers on the mthodological problems, shortcomings and flaws of network analysis would dispell some of the confidence in ts use.

    it would never stop its use by the anti-terrorism bureaucracies because they need some source of targets and network analysis provides that source.

    in any event, if the cia/dod use this science they ought at least to be doing extensive evaluation studies in order to move beyond “we think (hope) it’s the tool we need”.

    in a way, latching onto another set of social science theories is like believing spy satellites, invisible b-2 bombers, smart bombs, and now drones, will stop those suckers dead in their tracks – so to speak.

  12. orionATL says:

    @Peterr:

    “… People function within various social networks, which may or may not overlap. Killing someone removes them from ALL these networks, not only the AQ network. This person’s death may leave the AQ network confused and broken in terms of their functionality, but as Marcy notes, it potentially leaves other networks angry where they were neutral in the past. Thus, members of these second and third networks now may be inclined to become active members of the first, including becoming suicide bombers…”

    i would add this very grave potential loss:

    you are killing the natural leaders when you kill the al-q leaders. some of this may be unavoidable,

    but just to give an example, what a loss to the prospects for peace in the transition period as well as to the peoples and the nation of south africa if the apartheid goverment had executed nelson mandela.

  13. earlofhuntingdon says:

    Who says the US would limit digital “personality” targeting to active participants in generally recognized terrorist organizations? If we worked hard to derail British and Australian governments in the 1970’s, why would we not target (for political, not military assault) their contemporaries across the globe?

    I wonder how effective our “targeted killing” (assassination) program(s) would be if our opponents acquired the ability to construct virtual personalities, whose characteristics mimicked those the US is targeting, while recharacterizing the activities of their leadership into the “do not disturb” mode.

    I wonder if US actors are doing that to each other so as to avoid similar “targeting” (presumably for political assault) by state actors not enamored of US hegemony.

    As for the “human” side of a purportedly mixed human-s/w based targeting program, I’d be curious about the level of human involvement. Is it clerical and systems maintenance, with principal decision making delegated to the s/w component? Or does s/w troll for possibles, which are confirmed by humans who also make all controlling decisions about what to do then? As always, the devil’s in the details.

Comments are closed.